Richard F. Yalch (1975) ,"The Prediction of Voter Turnout", in SV - Broadening the Concept of Consumer Behavior, eds. Gerald Zaltman and Brian Sternthal, Cincinnati, OH : Association for Consumer Research, Pages: 117-130.
Broadening the Concept of Consumer Behavior, 1975 Pages 117-130 THE PREDICTION OF VOTER TURNOUT Richard F. Yalch, Assistant Professor, University of Washington The successful adoption of the marketing concept by businesses has stimulated nonprofit organizations to redefine their objectives in terms of exchanges with various publics. For example, a state university's objectives may be specified as educating a segment of the population, maintaining an intellectual climate for the faculty, supporting basic research to improve the public welfare, and convincing the legislature that state funds are being well spent. Each of these publics, students, faculty, citizens, and legislators, may be considered to be consumers of the university's products. This broad concept of marketing requires a broadened concept of consumer behavior and places a new burden on consumer behavior researchers. Previously, they had concerned themselves with the generality of a research finding from one product class to another. Now, it is proper to ask whether the postulated decision process explains behavior in situations in which the product may be a person, organization, or idea, and the price is one's vote, time, or allegiance. If the domain of consumer behavior is broadened to include noncommercial exchanges, concepts will have to be specified at a greater level of abstraction. The borrowing from the behavioral sciences that has characterized consumer behavior research to date can be more direct, allowing for an easier integration of psychological, sociological, economic, and political science theories. Moreover, by forcing a concentration on the stable, cross-situational causes of behavior rather than the unstable, situational adjustment variables, the behavior prediction problem can be redefined. The evidence that individual behavior in a specific situation is not perfectly predicted from a knowledge of the individual's attitudes (Wicker, 1969) takes on less importance. Rather, the effort should be to identify variables that reliably predict behavior across a variety of settings. The behavior prediction problem is thus not to develop a model that explains all of the variance in behavior but to determine which are the critical causal factors. The development of eclectic models to predict and explain behavior may characterize consumer behavior research in the "broadened era," and may represent its contribution to the other behavioral sciences. Consumer behavior research can benefit from the growing diversity in relevant situations because certain settings are more amenable to research than others. For example, comprehensive model testing has been restricted to the use of consumer panel data (Farley and Ring, 1971). However, marketers have been long concerned with the accuracy of self-reported purchasing data. Many nonbusiness settings readily provide actual behavior measures. Since voting, inoculation shot taking, and medical examinations usually require a recording of the subject's name and address, they might be more appropriate for testing a model purported to explain consumer choices. There are other methodological advantages to using a behavior such as voting as the subject of consumer behavior research. It occurs at fixed interval time points, has a well-defined consumer market in the registered voters, and must be purchased and used by a single individual. Thus, the problem of identifying when the consumer must make a choice, who really constitutes the market for the product, and incorporating the cases in which one party makes the purchase and another uses the product are alleviated. Moreover, because elections occur at least once per year, it is possible to take repeated behavior measures to test for the influence of past behavior and to average out the situational peculiarities that tend to reduce prediction accuracy. The methodological advantages associated with voter behavior research, which have made it a popular setting for research in the behavioral sciences, are not the only justification for using it for consumer behavior research. There is a generic similarity in the behaviors. Both involve primary (turnout) and secondary (candidate choice) demand decisions. Political parties must encourage their supporters to vote as well as convince them to vote for a slate of candidates, much as businesses try to promote the use of the product as well as their particular brand. The nature of the political campaign reflects this similarity. There is an initial attempt to build interest in the election and to develop name awareness, followed by an informational campaign to win converts, and usually ending with an emotional appeal to party loyalty. Therefore, the study of individual behavior during an election campaign should provide insights regarding how consumers respond to new product introduction campaigns. A voter's behavior actually consists of three distinct though related decisions, registration, turnout, and candidate choice. The registration decision is frequently independent of the election because once one has registered he is usually eligible to vote-for all subsequent elections. The major exception to this occurs when one has moved into a different district or has failed to vote for several consecutive elections. The turnout and candidate choice decisions are dependent on the election and the campaigns. These decisions are hierarchical as one cannot vote for a candidate until a ballot has been obtained. However, the decisions are not independent as a strong liking or 'disliking of a candidate may be the major motivation for going to the poll. The study presented here investigates the prediction of voter turnout problem rather than the candidate choice decision because the former can be unobtrusively assessed at the individual level. Political parties regularly maintain lists identifying voters and nonvoters to assist their election day campaign activities. In contrast, actual candidate choice can only be determined using aggregate data, the smallest unit typically being the precinct. This means that individual differences-in behavior, which serve as the basis for developing influence strategies, cannot be identified and aggregate data mask many of the important factors underlying behavior. Political science research of individual voter turnout is typified by the Survey Research Center's (SRC) national surveys. Their model considers both individual and election factors as predictor variables. The voters' feelings of political efficacy and citizen duty represent relatively stable qualities influencing turnout independent of the particular election. In addition, individuals are assumed to be exposed to campaign literature and mass media coverage which heightens their interest and involvement. As citizens become more aligned with a particular candidate or party, they are motivated to go to the polls to consummate their allegiance. If one is not attracted to a candidate, there is a low probability of his voting in the election. The SRC model can be reformulated as a two variable model, a stable disposition toward politics and a preference for a candidate. The former is usually assessed by self-reported political activity and the latter by attitudes toward the candidates and the election. There are costs associated with voting as well as the benefit of electing a desired candidate. Downs (1957) proposed that individuals evaluate the costs, the time and effort to become well-informed and to go to the polls, relative to the benefits, which includes demonstrating one's concern about the government. The information cost variable can be assessed as the voter's knowledge about the candidates and the election issues. The more informed the citizen is,, the more confident he should be that he will make the correct candidate choice. The benefit of voting is determined from the voter's attitudes and beliefs about the candidates, much as in the SRC model except that it is the perceived difference in the candidates that is the motivation to vote. Riker and Ordeshook (1968) extended Downs' model by incorporating another explanatory variable, the entertainment obtained from voting. They postulated that a voter might gain satisfaction from merely being a part of a major event. A voter's interest in the election can be considered to be a measure of the affective component of his attitude toward voting. Empirical research testing the voter turnout models is limited but demonstrates that the proposed variables are useful predictors. Shaffer (1972) tested the SRC and Downs models and found neither to be an adequate predictor of who would or would not vote in the 1964 Presidential election. only two variables were useful predictors, general orientation toward political involvement and attitudes toward the candidates. Silver (1973) also used SRC survey data to test turnout models, but concentrated on several cost variables. However, the results did not support his contention that differences in the cost of voting would account for the turnout variance. Instead, a noncost variable) interest in the election, was the best predictor. Although the voter behavior models have not accurately predicted turnout, several variables have been significant predictors. It would now seem appropriate to adopt the current attitude-behavior prediction research strategy of selecting a set of predictors drawn from different theoretical paradigms to construct a general prediction model. Variables such as normative beliefs (Fishbein, 1967), and the judged influence of extraneous events (Wicker, 1971) have reliably predicted behavior in a variety of situations. Their appropriateness to the voter turnout decision problem has yet to be determined. This study is designed to test the variables suggested as relevant to the prediction of voting behavior. Three hypotheses can be derived from the political science and attitude-behavior literature. First, attitudes toward the candidates are expected to be poor predictors of voter turnout. Fishbein (1967) distinguishes between attitudes toward the behavior and attitudes toward the object of the behavior. Only the former is expected to be a valid predictor of subsequent behavior because it better incorporates the situational adjustment that most directly determines behavior. Since there are a large number of ways that one can express a favorable or unfavorable attitude toward a candidate, it is unlikely that a particular measure will provide an accurate index of the individual's disposition. Therefore, attitudes toward the candidates should not be good predictors of turnout. The confirmation of this hypothesis would also imply that the turnout decision is independent of the candidate choice decision. The second hypothesis is that the "other variables" will be significant turnout predictors. Those tested are attitudes toward the election (affect toward voting), the judged influence of extraneous events, election knowledge, general political involvement, and behavioral intentions. The latter is expected to be the best predictor since it has been suggested as the immediate antecedent of behavior. The final hypothesis is derived from attribution theory's proposition that behavior determines one's attitudes (Bem, 1972), If one's past behavior is consistent, his attitude should be stable enough to be a predictor of future behavior. This is tested by assessing the voter's perception of the frequency of his voting in prior elections and determining how accurately it predicts turnout in the following election. THE MODEL A survey was conducted during a special aldermanic election in Chicago. The incumbent had resigned and five candidates filed nominating petitions for the nonpartisan election. The two candidates receiving the most votes participated in a run-off election one month later. In the week preceding this election, two samples were randomly selected from a list of registered voters. The first sample of 239 persons was interviewed by telephone, while the other 190 persons were contacted door-to-door. Paid interviewers administered the same questionnaire to both samples. The actual voting behavior of those interviewed was assessed from records kept by one of the political organizations supporting a candidate. In addition, most subjects were recontacted after the election for a short telephone interview and a self-reported voting measure was administered. Predictor Variables Attitude toward the candidates (A-c). This variable was computed in three different ways. First, the voter's attitude toward each candidate was determined using an expectancy-value formulation (Fishbein, 1967). Att1 = where,
A typical belief question was:
(2) very likely, (1) likely, (0) neither likely nor unlikely, (-l) unlikely, (-2) very unlikely The numbers in parenthesis represent the scoring for the possible responses.
A typical evaluation of beliefs (a1) question was:
(2) very desirable, (1) desirable, (0) neither desirable nor undesirable, (-I) undesirable, (-2) very undesirable The overall attitude toward each candidate was computed by summing the product of each belief and evaluation of belief response. For example, if one thought that it was unlikely that Bernard Stone would be an independent alderman (-1) and that it was very desirable to have an independent alderman (2), the product (-2) would be added to product for the other three issue questions to determine the voter's attitude toward Berm rd Stone. The attitude scores could range from +16 to -16. The voter's attitude toward the candidate was computed in the following three ways. "1-c -max (Att1, Att2 ). This represents the motivation caused by a strong liking of one candidate. "2-c = min (Att1 Att2). This represents the motivation that might arise from a strong disliking of one candidate. "3-c = Si (ai) . abs(B1i - B2i). This computes the voter's perception of the difference between the two candidates. The greater the perceived difference, the more likely he is to vote. Attitude toward the election (A-e). The variable provides a measure of the respondent's affection for the voting situation, as distinct from his attitude toward the objects of the behavior. It was formed by summing the responses to two questions.
(4) very interested, (3) somewhat interested, (2) not too interested, (1) not at all interested.
(4) care very much, (3) care somewhat, (2) don't care too much, (1) don't care at all. Social Normative Beliefs (NBs). This variable was employed to assess the individual's perception of what significant other persons think he should do in a given situation. It was operationalized in two questions:
(4) very important, (3) somewhat important, (2) not too important, (1) not important at all.
(4) very important, (3) somewhat important, (2) not too important, (1) not important at all. Judged Influence of Extraneous Events (EE) Scenarios describing four possible constraints on one's ability to vote in ail election were used to assess the strength of the voter's desire to vote.
For each of these questions, the following response categories were provided: (5) very likely, (4) likely, (3) neither likely nor unlikely, (2) unlikely, (1) very unlikely. Election Knowledge (K). Three questions were used to assess the voter's knowledge about the election.
The voter was given one point for naming each candidate and one point for knowing the correct endorsements, so the scale ranged from 4 to 0. General Political Involvement_(Inv). This variable represents the voter's long-term involvement in political affairs and was measured by four questions.
For each of these questions, the following response categories were provided: (3) very often, (2) often, (1) not too often, (0) never Voting Frequency (VF). The subject's self-perception as a frequent or infrequent voter was determined by asking how often he votes in different types of elections.
Self-reported Past Behavior (PB). During the pre-election interview, the voter was asked if he had voted in the preliminary election held three weeks earlier.
Behavioral Intentions (BI). The voter's subjective probability of voting in the run-off election was assessed by the question:
(7) definitely will vote, (6) probably will vote, (5) might vote, (4) unsure, (3) might not vote, (2) probably will not vote, (1) definitely will not vote. Criterion Measure Actual July Voting (B). An unobtrusive measure of the subject's behavior on election day was obtained from party records. The advantage of this method was demonstrated by the ten per cent error in a self-reported voting question included in the post-election interview.RESULTS The means and variances of the two independent samples were compared and since no differences were significant, the data was pooled to provide an overall sample of 429. However, this was reduced to 374 for the multivariate analysis because of missing data. Table 1 presents the correlations between the independent variables and the point biserial correlations between these variables and the two dichotomous behavior measures. Simple Correlations The best predictors of voting in the July run-off election are attitudes toward the election (A-e), behavioral intentions (BI), the voter's perception of how frequently he votes (VF), and his past behavior (PB). other variables with a correlation greater than .20 are the judged influence of extraneous events (EE) and election knowledge (K). As predicted, attitudes toward the candidates do not appear to explain very much of the turnout variance. Furthermore, the SRC political involvement measure (Inv), which has proven to be a useful predictor in national election surveys, is insignificantly associated with behavior. The advantage of using an unobtrusively monitored behavior rather than a retrospective self-report is demonstrated by the difference in the biserial correlations. Predictors,, such as the judged influence of extraneous events (EE) have a much greater association with a criterion measure assessed in the same interview, reflecting the influence from sharing the same measurement technique. If only the retrospective data has been collected,, one might have overestimated these variables' importance. TABLE 1 CORRELATION MATRIX Logit Analysis A multivariate analysis was performed to determine the combined effect of the variables. Since the dependent variables is dichotomous, ordinary least squares regression is not appropriate. Therefore, logit analysis was used (Nerlove and Press, 1973). It provides maximum likelihood estimates of each independent variable's effect on an individual's voting probability. These estimators are normally distributed and their significance may be tested using a student's t-test. Table 2 presents a summary of the results for 14 different models. The Chi-square represents the overall fit of a model relative to a simple binomial model which assumes that all the coefficients are equal to zero. Thus , an individual's voting probability is assumed to be equal to the proportion voting in the total population. The F-statistic is used to measure prediction accuracy. It is computed by grouping the predicted voting probabilities into ten intervals (for example, 0-.1, .1-.2, ...), and comparing the mean predicted probability with the actual proportion voting. Because we are primarily interested in explaining why individuals decide to vote, the Chi-square will be used to select the best model. The first model in Table-2 tests the effect of all variables. [Models using social normative beliefs and the attitude toward the candidates are not presented because their coefficients had + values that were consistently less than 1.0.] Only attitude toward the election (A-e), general political involvement (Inv), and behavioral intentions (BI) have coefficients significantly different from zero (t> 1.96, p /,.05). The unexpected finding that Inv is a critical variable in the multivariate analysis but not in the simple correlation (see Table 1), is an apparent consequence of its serving as a "suppressor" variable (Wiggins, 1973, pp. 30-35). It removes from the regression equation a portion of the variance in the independent variables that is irrelevant to the variance in the criterion variable. The suppressor effect is indicated by the negative sign on tie Inv. coefficient. this statistical finding as representing the influence. One might interpret of the political attitude variables on nonvoting political behaviors. By removing that relationship from the model, the attitude-votitig behavior relationship becomes better isolated. The second model drops BI and PB from the analysis to provide an estimate of the explanation provided by the disposition variables. Although only A-e and Inv have significant coefficients, the coefficients for the other variables have the correct signs and approach significance. Models III through VIII represent different combinations of the independent variables. It is apparent that the underlying dispositions measured by these variables are very similar, if not identical. As the number of predictor variables is reduced from five to three, there is only a minor decrease in the Chi-square (from 60.0 to 55.3). However, if all three marginal variables (VF, K, and EE), are dropped, the Chi-square drops to 50.3, which is a significant change (p=.02). Models XIII and XIV represent parsimonious prediction models and demonstrate the importance of prior behavior and a very specific disposition toward voting in this election on actual voting. The latter result is congenial with Fishbein's (1967) proposition that the more specific an attitude measure is with regard to a future behavior, the closer the association. Nevertheless, in all fourteen models, a major portion of the variance remains unexplained. TABLE 2 LOGIT ANALYSIS OF MULTIVARIATE MODELS DISCUSSION The results support the hypothesis that the variables selected would be significantly associated with actual voting behavior. Although none of the prominent voter turnout models is explicitly tested, it does appear that the approach of selecting variables from several models, including general attitude-behavior prediction research enhanced prediction accuracy. This is evidenced by the statistically significant contribution of the judged influence of extraneous events and self-perceived voting frequency to the variance explained by the standard voting variables, such as attitudes toward the election and the candidates. Moreover, the search for a universal set of unique-predictors did not predict better than several three variable models can be attributed to the redundancy in the dispositions being assessed. The development of more orthogonal variables is needed to advance the effort to create a cross-situational consumer behavior model. The value of individual behavior prediction models in suggesting change strategies is readily demonstrated. For example, attitudes toward the specific election was the best predictor. Therefore, the manipulation of its determinants, election interest and involvement, could be used to influence turnout. A campaign manager might allocate door-to-door canvassers and direct mailings by districts according to a priori preference estimates such as to stimulate turnout in favorable districts and avoid stimulating it in unfavorable districts. Other significant associations suggest additional campaign strategies. The effect of the judged influence of extraneous events measure indicates the importance of both minimizing the external constraints on voters (transportation to and from the polling place, protection from bad weather, sitters for the children,, etc.), and convincing supporters that the benefits of voting exceed these costs. Since it does not appear that favorable attitudes toward the candidates stimulated turnout, it might be more advisable to concentrate on promoting the desirability of citizen involvement and convincing individuals that voting takes little effort. A third important variable is election knowledge. Voters who are not familiar with the candidates and the issues are less likely to vote. The typical procedure for manipulating this factor is to flood the community with campaign propaganda. This practice may be questioned as it is possible that the voters may begin to feel that the choice decision is too complicated. That is, while they might be acquiring more information, they might also be learning that the total amount available is very great and conclude that they are not competent to make a satisfactory choice. On the other hand, if only one or two issues were identified and the campaign organized around the candidate differences on these issues, the voters would more readily perceive themselves as having sufficient knowledge. Elections that have focused on simple salient issues have stimulated high turnout. A good example is the 1972 George Wallace primary campaign in Michigan. His campaign theme, "Send Them a Message," tied in nicely with the controversy over school bussing and voter turnout exceeded expectations. The fourth significant variable is the voter's self-perception of his typical election response. Individuals who viewed themselves as usually voting were more likely to actually vote. The tendency to behave in accord with one's self-concept might also influence the candidate choice decision. Candidates often attempt to broaden their appeal by creating peripheral organizations to publicize their diverse support. In the 1972 Presidential election, Democrats for Nixon and Businessmen for McGovern were established, probably not so much to raise money and disseminate information but to reduce the inhibition some voters felt about supporting either candidate. Because of the promotion of the belief that there was nothing incongruent about being a member of one party and a voter for the candidate of the other party, Nixon attracted many Democratic votes and had a landslide victory. The final important predictor is voting in the most recent election, a result which is consistent with self-perception theory's proposition that past behavior determines one=s attitudes toward an object. Marketers have developed techniques to encourage the initial trial of new products in the belief that this experience will lead to repeated purchasing. However, the importance of the circumstances surrounding the initial trial has not been adequately researched. Deci (1971) suggests that even enjoyable behaviors may be discouraged by their continued association with external inducements. Campaign managers should consider strategies designed to induce positive behaviors toward the election and their candidates without these inducements becoming perceived as the sole justification for this, behavior. The advantage of low pressure and no pressure compliance has been demonstrated in nonvoting contexts (Freedman and Fraser, 1966; Scott, this volume). The significant associations between the set of predictors and actual voting behavior is, of course, only suggestive of causality. Experimentation is required to confirm them. However, this discussion does illustrate the usefulness of individual behavior prediction research as a method to develop change strategies. This study also supports the proposition that an eclectic approach is needed to properly research consumer behavior in the broader contexts in which it is now recognized as occurring. REFERENCES Bem, D. J. "Self-Perception Theory," in L. Berkowitz (ed.), Advances in Experimental Social Psychology. New York: Academic Press, 1972. Brislin, R. W. and Olmstead, K. H. "An Examination of Two Models Designed to Predict Behavior from Attitude and Other Verbal Measures," Paper presented at the American Psychological Association Meeting, Montreal, Canada, 1973. Campbell, A., Converse, P., Miller, W. E. and Stokes, D. The American Voter. New York: John Wiley and Sons, 1960. Deci, E. L. "Effects of Extrinsic Mediated Rewards on Intrinsic Motivation," Journal of Personality and Social Psychology, 1971, 18, 105-115. Downs, A. An Economic Theory of Democracy. New York: Harper and Row, 1957. Farley, J. U. and Ring. W. "'Empirical' Specification of a Buyer Behavior Model," Journal of Marketing Research, 1974, 1-1, 89-96. Fishbein, M. "Attitudes and the Prediction of Behavior," in M. Fishbein (ed.), Readings in Attitude Theory and Measurement. New York: John Wiley, 1967. Freedman, J. and Fraser, S. "Compliance Without Pressure: The Foot-in-the-door Technique," Journal of Personality and Social Psychology, 1966, 4, 195-202. Nerlove, M. and Press, S. J. Univariate and Multivariate Log-linear and Logistic Models. Santa Monica, California: The RAND Corporation, December, 1973. Riker, W. H. and Ordeshook, P.C. "A Theory of the Calculus of Voting," American Political Science Review, 1968, 62, 25-42. Scott, C. A. To be titled and assigned pages in this volume. Shaffer, M. "A Demand Analysis of Voting Costs and Voting Participation," Social Science Research, 1973, 2, 111-124. Wicker, A. W. "Attitudes Versus Actions: The Relationship of Verbal and Overt Behavioral Inconsistency," Journal of Personality and Social Psychology, 1971, 19, 41-78. Wicker, A. W. "An Examination of the 'Other Variables' Explanation of Attitude-Behavior Inconsistency," Journal of Personality and Social Psychology, 1971, 19, 18-30. Wiggins, J. S. Personality and Prediction: Principles of Personality Assessment. Reading, Mass.: Addison-Wesley, 1973. ---------------------------------------- |
What is the single best predictor of vote choice
Postagens relacionadas
direito autoral © 2024 mesadeestudo Inc.
